Previous conference records:

• 230925_3rd_DDS@Osaka
• 230926_3rd_DDS@Osaka

Conference Records 2024-01-18:

Interesting sessions

Session name: Bioscience Group

Details and Comments

• Presented study simulated the polarization of P1 cell
• Tested 2 proteins and found the trigger of the polarization.
• Forwards to model the dynamics of cortical flow.

Session name: Data Science Group

Details and Comments

Need to know (dynamics structure):

• Koopman operator and Dynamic Mode Decomposition
• Method: Discrete eigen convert to continuous spectra (Residual DMD)

Need to know (generative models):

• Diffusion models and its core: reverse SDE
• Flow matching: ODE

Comments:

• 1st focused on clustering spectra,excluding the noise.
• 2nd part of the presentation might be good at application

Session name: Recruit Session 01

Details and Comments

• Experiments: $M{g}_{94}Z{n}_2{Y}_4$ as object
• Machine Learning: Voxel density → self-correlation → LPSO-LPSO → persistent index
• Reverse analysis (PH reverse): simulate the original structure of the material to find the gaps

Session name: Recruit Session 02

Details and Comments

• on factor analysis: $X=L\psi +B$
• $\psi =0$ or negative, lead to improper question
• New plan: computational method rather than heuristic

Session name: Recruit Session 03

Details and Comments

• Focus on outlier of persistent homology
• Approach: modify the radius (density probability distribution → find outlier)

Session name: Recruit Session 04

Details and Comments

• need to know: Fisher-Rao metrics

Session name: Recruit Session 05

Details and Comments

• Constructing differential equation using time-series data
• Observations → model coordinates → estimated time derivatives → system
• ODE Estimation: modeling from time series

Session name: Recruit Session 06

Details and Comments

• Observations of real wind speed, direction
• Using TDA to predict rain with wind speed and wind direction
• Plot the change of persistent homology to predict probability of rain

Session name: Application & Exploration Group

Details and Comments

• Data: High-dimensional dataset
• Distance: $d(i,j)=(1-y_i\ast y_j{)}^{1/2}$
• Hypothesis: Crisis in market lead change of topological index

Session name: Fundamental Research Group

Details and Comments

• Determinantal point processes (DPPs)
• Need to know: Hyperuniformity
• Session 2: Topology smoothing (points → estimation → PD)
• Solution: Graph theory $Q_n^{ϵ}:=T_0^{ϵ}{T}_1^{ϵ}{T}_2^{ϵ}...$

Daily Summary

What's new (for me)

• Flow matching method and ODE
• Fisher-Rao metrics

What should I check

References

Dao et al., 2023, Flow Matching in Latent Space https://arxiv.org/abs/2307.08698
Lipman et al., 2023, Flow Matching for Generative Modeling https://arxiv.org/abs/2210.02747
Mayama et al., 2022, α-Mg/LPSO (Long-Period Stacking Ordered) phase interfaces as obstacles against dislocation slip in as-cast Mg-Zn-Y alloys https://www.sciencedirect.com/science/article/pii/S0749641922000766